基于启发式广度优先搜索的SVR参数优化方法研究

Study on Parameters Optimization of Support Vector Regression Based on Heuristic Breadth-first Search Algorithm

支持向量回归机(Support vector regressio,SVR)模型的拟合精度和泛化能力取决于其相关参数的选择,其参数选择实质上是一个优化搜索过程。根据启发式广度优先搜索(Heuristic Breadth first Search,HBFS)算法在求解优化问题上高效的特点,提出了一种以k-fold交叉验证的最小化误差为目标,HBFS为寻优策略的SVR参数选择方法,通过3个基准数据集对该模型进行了仿真实验,结果表明该方法在保证预测精度前提下,大幅度的缩短了训练建模时间,为大样本的SVR参数选择提供了一种新的有效解决方案。

The regression accuracy and generalization performance of support vector regression（SVR） models depend on a proper setting of its parameters,but parameters selection is an optimization problem.Motivated by the characteristic of heuristic breadth first search（HBFS） on optimization problem,a new automatic searching methodology based on HBFS algorithm is proposed in this paper.In this method,k-fold cross-validation error is used as the fitness function of HBFS,Results of 3 benchmark datasets show that the new method not only can assure the prediction precision but also can reduce training time markedly.The new method is an efficient solution to large-scale samples model optimization for SVR.